def _str_to_relative_instruction()

in tensorflow_datasets/core/splits.py [0:0]

• 47 lines of code
• 19 McCabe index (conditional complexity)

def _str_to_relative_instruction(spec: str) -> AbstractSplit:
  """Returns ReadInstruction for given string."""
  # <split_name>[<split_selector>] (e.g. `train[54%:]`)
  res = _SUB_SPEC_RE.match(spec)
  err_msg = (f'Unrecognized split format: {spec!r}. See format at '
             'https://www.tensorflow.org/datasets/splits')
  if not res:
    raise ValueError(err_msg)
  split_name = res.group('split_name')
  split_selector = res.group('split_selector')

  if split_name == 'all':
    if split_selector:
      # TODO(tfds): `all[:75%]` could be supported by creating a
      # `_SliceSplit(split, from_=, to=, unit=)`.
      raise NotImplementedError(
          f'{split_name!r} does not support slice. Please open a github issue '
          'if you need this feature.')
    return _SplitAll()

  if split_selector is None:  # split='train'
    from_ = None
    to = None
    unit = 'abs'
  else:  # split='train[x:y]' or split='train[x]'
    slices = [_SLICE_RE.match(x) for x in split_selector.split(':')]
    # Make sure all slices are valid, and at least one is not empty
    if not all(slices) or not any(x.group(0) for x in slices):
      raise ValueError(err_msg)
    if len(slices) == 1:
      from_match, = slices
      from_ = from_match['val']
      to = int(from_) + 1
      unit = from_match['unit'] or 'abs'
      if unit != 'shard':
        raise ValueError('Absolute or percent only support slice syntax.')
    elif len(slices) == 2:
      from_match, to_match = slices
      from_ = from_match['val']
      to = to_match['val']
      unit = from_match['unit'] or to_match['unit'] or 'abs'
    else:
      raise ValueError(err_msg)

  if from_ is not None:
    from_ = int(from_)
  if to is not None:
    to = int(to)

  return ReadInstruction(
      split_name=split_name,
      rounding='closest',
      from_=from_,
      to=to,
      unit=unit,
  )